ABSTRACT Skin is the largest organ in the human body and covers the entire body. Face skin diseases are a wide range of disorders that can affect the skin on the face. In this paper, a novel S3-GHOSTNET has been proposed to identify the face skin diseases. Initially, the input images are gathered from the Derm Net dataset and pre-processed using the Contrast stretching Adaptive Gaussian star Filter for enhancing the quality of the image. The enhanced images are fed into Modified Deep V3+ for segmenting the diseases affected region in the images. Then the segmented images are fed into deep learning-based Shuffle Net for extracting the features of the images. Finally, the required features are selected using Sand Cat Swap Optimization and using Ghost Net the Penta type of face skin diseases are classified. The proposed method achieves an accuracy rate of 99.34% in the normal classes and 98.94% in the abnormal classes. Additionally, the proposed method archives high-level Precision, Recall and F1 Scores of 98.05%, 97.73%, and 97.66% in the normal and abnormal classes. The proposed S3-GHOSTNET achieve the overall accuracy rate of 0.49%, 1.49%, 0.36% and 0.79% comparing the existing methods such as GLCN, Leaky RELU, CNN and Eff2Net, respectively.
Read full abstract